We propose to develop, apply, and test a novel algorithm to retrieve seamless and consistent, sub-hectare
resolution estimates of land cover variables from multiple satellite data sources (e.g., Landsat, Sentinel-2,
and ALOS-PALSAR) at continental to global scales. Using a generalized approach to estimation that
enables assimilation of multiple data sources, this approach has generated the world's first Landsat-based
percent tree cover dataset; an annual time-series of urban impervious cover from 1985 to 2010; a Landsat-based,
circa-1990 map of global forest cover and change; and a global surface-water layer. Uniquely, the approach
estimates both cover and its uncertainty at the scale of pixels, a feature on which the rigor of analyses
such as change detection depends.
Our existing algorithms will be adapted to produce prototype percent-tree and water-cover layers globally in
2000, 2005, and 2010, as well as continentally -over North and South America- at annual frequency from 2010 to
2015 from passive-optical (Landsat and Sentinel-2) and SAR measurements. Generating a global dataset at annual
frequency is beyond the scope of this proposal; however, North and South America represent all of the world's
major biomes and so offer the complete global range of environmental sources of error and uncertainty.
Development of data layers at these two scales will demonstrate the efficient use and seamless combination of
multiple data sources and will develop the capability for periodic satellite-based inventories of land-cover
and land-use change using NASA computing facilities. The resulting layers will be validated against highly
accurate, independent reference data, and all estimates will be accompanied by global and local (i.e., per-pixel)
estimates of their uncertainty. This research will be partnered with the ESA-funded GLOBBIOMASS project
(C. Schmullius, PI), which will receive our estimates of tree cover to map biomass regionally and
globally in 2005, 2010, and 2015 epochs.